Implementing GANs to Recognize Handwritten Digits

In this chapter, we will build an Android application that detects handwritten numbers and works out what the number is by using adversarial learning. We will use the Modified National Institute of Standards and Technology (MNIST) dataset for digit classification. We will also look into the basics of Generative Adversarial Networks (GANs).

In this chapter, we will take a closer look at the following topics:

• Introduction to GANs
• Understanding the MNIST database
• Building the TensorFlow model
• Building the Android application

The code for this application can be found at https://github.com/intrepidkarthi/AImobileapps.

Introduction to GANs

GANs are a class of machine learning (ML) algorithm that's used in unsupervised ML. They are comprised of two deep neural networks that are competing against each other (so it is termed as adversarial). GANs were introduced at the University of Montreal in 2014 by Ian Goodfellow and other researchers, including Yoshua Bengio.

Ian Goodfellow's paper on GANs can be found at https://arxiv.org/abs/1406.2661.

GANs have the potential to mimic any data. This means that GANs can be trained to create similar versions of any data, such as images, audio, or text. A simple workflow of a GAN is shown in the following diagram:

The workflow of the GAN will be explained in the following sections.

Generative versus discriminative algorithms

...

Summary

Using this Android application, we can learn how to write a freehand writing classifier using TensorFlow Lite. With more data on handwritten alphabet datasets, we should be able to identify alphabets in any language using GANs.

In the next chapter, we will build a model for sentiment analysis and build an app on top of it.